module RubyLLM::TestHelper

Lets a test script what the model replies, with no provider and no API key. Replies go through the real conversation loop, so streaming, tools, persistence, cost, and #cancel all behave as they do in production. Require it in your test helper, and include it like ActiveJob::TestHelper:

require "ruby_llm/test_helper"

class SupportAgentTest < ActiveSupport::TestCase
  include RubyLLM::TestHelper

  test "answers" do
    stub_replies "Hello there", cost: 0.01

    reply = SupportAgent.new.ask("Hi")

    assert_equal "Hello there", reply.content
    assert_equal 0.01, reply.cost.total
  end
end

It also works in RSpec: config.include RubyLLM::TestHelper. A chat’s request reaches the stub. Other operations, such as RubyLLM.embed, still call their provider.

Public Instance Methods

stub_replies (*replies, cost: nil, interval: nil)

Source

# File lib/ruby_llm/test_helper.rb, line 107
def stub_replies(*replies, cost: nil, interval: nil)
  Script.stub(replies, { cost:, interval: })
end

Makes every chat answer with replies until the test ends, in order, repeating the last, instead of calling a provider. A String streams word by word, and tool_call asks the conversation loop to run a tool. cost: is what each reply reports, in dollars. interval: is the seconds between words, so a test can press Stop while a reply streams.

stub_replies "Hello there", cost: 0.01
stub_replies tool_call(:weather, city: "Berlin"), "It is sunny."
stub_replies "A long reply", interval: 0.1

tool_call (name, **arguments)

Source

# File lib/ruby_llm/test_helper.rb, line 113
def tool_call(name, **arguments)
  ToolCall.new(id: "call_#{SecureRandom.hex(4)}", name: name.to_s, arguments:)
end

Returns a reply that calls the tool name with arguments, for stub_replies.